This proposal addresses Provocative Question #2. We will use innovative approaches to investigate how CDKN2A (which encodes p16) mutation carriers develop different cancer phenotypes (pancreatic cancer vs melanoma vs no cancer), and include both genetic and non-genetic factors. We have identified 4 large, multi- generational kindreds with a founder CDKN2A deleterious mutation (L16R, 47T>G). Our preliminary observations demonstrate that this mutant has lower expression and decreased ability to regulate cell cycle progression compared to wild type protein. Our sequencing studies of kindred members with different cancer phenotypes have identified potential variants in novel genes that modify risk (LGR6, a co-receptor of Wnt signaling and COL11A1, which participates in oncogenic signaling, including TGFbeta). We will determine the ability of the p16 mutant to promote transformation and how it is influenced by interaction with the above candidate modifier genes, LGR6 or COL11A1, in pancreatic cancer and melanoma. We will also develop novel computational models using machine deep learning, to generate networks that capture high dimensional features to integrate gene, biology, and cancer phenotype. This approach will be extended to kindreds with other CDKN2A mutations. Our Specific Aims are to: (1) Identify genotypes of potential modifier genes in multiple kindreds that feature pancreatic cancer and melanoma and known to carry CDKN2A germline mutations. We will use genome wide variant coverage of germline DNA from CDKN2A carriers from the 4 large L16R kindreds, plus additional members in 42 other similar CDKN2A kindreds. We will identify candidate modifier genes in the kindreds by rule-based statistical genetic analysis of genotypes. (2) Define the impact of CDKN2A L16R mutation on the function of p16 and its interplay with candidate modifier genes. We will elucidate the biological significance of mutations in CDKN2A and candidate modifier genes using functional and high throughput methodologies by analyzing the mechanism underlying the interplay between p16 and modifier genes; define new pathways cooperating with this interplay using a combination of genome wide studies to assess transformation in cells carrying p16 mutant or wild-type background using well established in vitro and in vivo models. (3) Develop a deep learning network model to integrate genetic, biological and epidemiological data to accurately infer pancreatic cancer and melanoma phenotypes and age of onset in mutation carriers. We will apply a convolutional neural network, a deep learning algorithm in the training dataset, develop a back-propagation algorithm to fine tune ?weights,? and construct mutation-gene networks to capture high-dimensional features for each disease subclass. We will acquire and disseminate new knowledge and tools to the scientific community. Our integrated methods and approach will bring insight into how different cancer phenotypes can occur with identical predisposing mutations, which can be applied to other cancer syndromes with similar challenges.